2022-06-07

Definitions

  • Installation Year: The year a failed part was put in place (not necessarily manufactured): usually grouped by decade in this analysis
    • manufactured vs install year is hard to compare since the former is rarely given
  • Decade-year: Referring to the rate of pipeline failure attributable to a given install decade in a specific incident year
    • i.e. The rate of 1960’s incidents in 2015
  • Incident Rate: Always given in incidents per 1,000 miles in this presentation
  • Period of Interest: 2012 to 2021, but we can go back to 2010
    • sticking to PHMSA’s timeline just for ease of comparison for now
  • In plots, PHMSA or PST U is for unweighted averages, while PHMSA or PST W is a weighted average

Aligning with PHMSA

GT Incident Characteristics:

  • 10 most recent complete calendar years (2012-2021)
  • Significant
  • Onshore Pipeline, including valve sites
gt.inc %>%
  mutate(INSTALLATION_YEAR = parse_number(INSTALLATION_YEAR,na = c("NA","UNKNOWN")),             
         INSTALLATION_DECADE = INSTALLATION_YEAR - INSTALLATION_YEAR %% 10,
         INSTALLATION_DECADE = if_else(INSTALLATION_DECADE <= 1930, 1930, INSTALLATION_DECADE),
         INSTALLATION_DECADE = replace_na(INSTALLATION_DECADE, 1920))%>%  
  filter(SIGNIFICANT == "YES",   
         ON_OFF_SHORE == "ONSHORE",  
         between(IYEAR, 2012,2021),
         SYSTEM_PART_INVOLVED == "ONSHORE PIPELINE, INCLUDING VALVE SITES" )

PHMSA’s Graph

My Graph

Misalignment

  • Approach to “averages”
    • Weighted mean takes into account changes in incidents counts and mileage
  • Weird math in PHMSA’s 2020s: SQL or other issues?
  • Inclusion of other incident parts and locations
    • Parts: compressor stations, storage, regulator/metering
    • Locations: offshore (platforms?)
  • Another consideration is potentially examining incidents by age of part at the time of incident
    • A little less clear when it comes to using mileage by decade

Arithmetic Mean

\[ \mu_d = \frac{\sum_{y = 2012}^{2021}Inc_{dy}}{\sum_{y = 2012}^{2021}Miles_{dy}} \]

  • \(Inc\) = Number of incidents in a given period
  • \(d\) = Decade of a given installation year (40s through 2020s)
  • \(y\) = Year incident occurred (2011-2021)
  • \(\mu\) = Average incident rate for a given \(d\)

Weighted Mean

How do we “weigh” the mean? By adjusting the raw incident rate in a given decade-year. Notice that this time \(\mu_{dy}\) is only for a specific decade-year (\(dy\)) and \(mu_d\) involves multiplying the sum of \(\mu_{dy}\) by the number of possible incident years it includes (10, except in the 2020’s).

\[ \begin{aligned} \mu_{dy} = \frac{Inc_{dy}}{Mileage_{dy}}* \frac{Inc_{dy}}{Inc_{d}}\\ \mu_d = \frac{\Sigma_{y=2012}^{2021}{\mu_{dy}}}{10}\\ \end{aligned} \]

What about the 2020s?

Why does PHMSA present an incident rate of around 1.4 in the 2020s? Below is a table of all incidents in the relevant period (2012-2021), relevant filters, and an install year in the 2020s.

Operator Location SIGNIFICANT Inc. Yr. Inst. Yr. Dec-Yr Mileage
NORTHERN NATURAL GAS CO Hardin County, IA YES 2,021 2,021 3,628


1 incident in a year with 3,628 miles offers a rate of .28 for the year. Due to a lack of 20-20 decade-year incidents, the weighted average is .56 while the unweighted is .14. Either way, a single incident in a single year is basically statistical noise, and inconclusive.

Adjusting the Filters

So how may incidents are we actually looking at?

Install Year
Part n Mean Median Mode
ABOVEGROUND STORAGE, I… 2 2002 2002 1982
BELOWGROUND STORAGE, I… 7 1971 1961 1982
OFFSHORE PIPELINE, INC… 74 1977 1974 1970
OFFSHORE PLATFORM, INC… 7 1986 1979 1997
ONSHORE COMPRESSOR STA… 148 1991 1998 2016
ONSHORE PIPELINE, INCL… 401 1966 1963 1959
ONSHORE REGULATOR/METE… 38 1989 1999 2010

In Short:

  • 677 in the time period we are looking at
  • 59% are currently included in PHMSA’s analysis
  • Offshore incidents (which we have mileage for, sort of) would increase that to 70% of incidents

Why Don’t I Have a Plot Including Offshore Incidents?

“User hostile” doesn’t even begin to touch how much more complicated the offshore mileage reporting is. Aside pipeline with OCSG, OCSP, and OCSAT in the State column, the per-decade mileage of offshore pipeline is slightly unclear and requires some investigation with PHMSA.

It seems like in some cases offshore and onshore mileage reporting is mixed, so why is PHMSA reporting rates for only onshore? I have no idea, but the more I investigate this topic the more confused I am about everything I’ve heard from them. Also, I have only just noticed the PHMSA column in the incident data that mentions OCS incidents, and just haven’t incorporated that yet.

What About Other Onshore Parts?

Installation Decades and Failed Parts

Description

This next plot is a bit confusing so needs a bit more introduction. It shows the percentage of significant incidents attributable to a specific part that was installed in a given decade. In other words, one of two incidents attributable to Aboveground Storage occurred on a storage part that was installed in the 1980s, thus the aboveground storage part registers 50% in the 1980s column (the other incident was attributable to AS part installed in the 2020s, so it’s 50% there as well). I used percentages here to account for the vastly different incident counts attributable to each part (from 2 to 401), although it is a bit less immediately intuitive. What I hope is visible from this is a relative increase in compressor station and onshore regulator incidents installed after the 1980s (green and orange dots).

Installation Decades and Failed Parts

Plot

Unknown?

Apparently, the mileage data download has separate columns for Onshore Pre-40s and Unknown mileage. What does it look like when we separate them?

Final Q’s

PHMSA

  • Is there a credible way to handle offshore mileage and incidents?
    • If not, why?
  • Why are they still presenting rates as incidents per 100 miles but calling it per 1,000?

PST

  • Hw do we handle non-pipeline & valve site incidents (compressor stations, storage, etc.)?
  • How do we handle “unknown” years/decades?
  • Does using 10 years vs all 12 matter?